You are on page 1of 3

FSR Forum | september 2009 | 

FSR Forum | september 2009 | 

Success is a team effort!

Trading on the future.
All Options is a leading market maker providing liquidity to the derivatives markets in Europe and Asia. In only 2 years we have grown from 60 to 300+ employees and are now one of the largest market makers in the world. And we dont stop there this year we are seeking another 50 young talents for trading careers. This success is due to our belief in support for personal achievement and discipline in success. Working in unity is at the core of our culture. Thats why we reward our traders on both their individual and team performance. If you are interested in a challenging and rewarding career in trading check out your options at:

Conditional estimation of linear asset pricing models using alternative marginal utility growth instruments1
By David J. Moore
The exchange economy model of Lucas yields the familiar Euler condition that discounts expected returns by marginal utility growth. Unfortunately, proxies for marginal utility growth are subject to theoretical and empirical shortcomings
1 Introduction

This study proposes a new macroeconomic theory-derived and productivity-based marginal utility growth proxy that avoids the theoretical difficulty in utility function specification and empirical problems associated with consumption data used as the primary utility function input. The theoretical and empirical characteristics of this proxy indicate it may be useful in linear asset pricing models as an instrumental variable.
2 Exchange economy model

sides by ptu [ct] and defining Rt+1 = (pt+1 + dt+1) / pt equation (1) can be expressed in discount factor form:

the discount factor coefficient of variation CV [m] = [m]=E[m] provided by Hansen and Jagannathan (1991):

The exchange economy model (Lucas, 1978) yields the Euler condition for a representative agents intertemporal utility maximization:

The stochastic discount factor (3) is unobservable therefore explicit utility function assumptions must be made before applying the model. Two frequently used utility functions, constant relative risk aversion (CRRA) utility and log utility are now used to illustrate theoretical and empirical difficulties associated with marginal utility growth proxies. CRRA utility CRRA utility can be expressed in the power-utility form u[c] = c1- / (1-) where , a constant, is the relative risk aversion coefficient. Taking the first derivative and inserting into (3):

where Et is the time t expectations operator, pt is the price at time t, ct is the consumption at time t, is the rate of time preference (subjective discount factor), and dt+1 are dividends paid at time t+1. The equality in (1) reveals the cost in marginal utility (ptu [ct]) of purchasing the asset must be equal to the discounted () expected gain of the future payoff (u t [ct+1] (pt+1 + dt+1)). By dividing both

This discount factor is at the heart of the equity premium puzzle identified by Mehra and Prescott (1985). The puzzle can be illustrated by the lower bound on

where Re is the return in excess of the Treasury Bill rate. The post-war Sharpe ratio (E [Re] / [Re]) for the valueweighted NYSE portfolio is about 0.5 while per-capita consumption growth has a standard deviation of about 1% per year (Cochrane, 2005). Given the discount factor of equation (4), and setting = 1, the required coefficient of risk aversion to satisfy (5) is approximately 50. However, Mehra (2003) noted a large body of literature that suggests the coefficient of risk aversion is certainly less than 10. The extremely high risk aversion indicates the volatility of consumption growth is too low to explain the equity premium. CRRA utility does not fit observed data without implausibly high levels of risk aversion, but what about other utility function forms? Other more complex utility functions have been proposed such as habit formation and time-nonseparable forms. Unfortunately, the more complex forms have proven unsuccessful in resolving the equity premium puzzle >>>

Join our team:

)E  arlier versions of this paper were presented at Pepperdine University, The University of Memphis, The University of Tennessee, Lynchburg College, and during the 2009 Ph.D. Project Finance Doctoral Student Association meetings coincident with the 2009 Western Finance Association annual meeting.

FSR Forum | september 2009 | 

FSR Forum | september 2009 | 

Figure 1: Graphical representation of general equilibrium macro growth model

tivity shock, and Xt is the deterministic component of productivity which grows at a constant (and exogenous) rate > 1:

Solution; The first order conditions of the maximization problem are:

Table 1: Data Items, sources, and availability

I am not addressing labor-leisure choice in this study; therefore, labor is assumed to be fixed (Nt = 1t) and the production function simplifies to: The first order conditions (17) and (18) can be combined to reveal the key contribution of this study: an alternative proxy for marginal utility growth:

as well (Mehra, 2003). From log utility to CAPM Log utility u [ct] = ln [ct] can be shown to map directly into CAPM. Let Rtw represent the return to the market wealth portfolio with the S&P 500 return often used as an empirical proxy. Cochrane (2005) found that given log utility,

coefficients. Cochrane also noted these time-varying coefficients account for time-varying expected returns, variances, covariances, and risk free rates. Here, I provide an additional perspective: the time varying coefficients of (8) also account for time-varying risk aversion, specifically fluctuations about = 1 given the connection between log-utility and CAPM.
3 Macroeconomic growth model

Since mt+1 =u [ct+1] = u [ct], Rtw+1 = 1/mt+1 or mt+1 = 1/Rtw+1. Taking the linear approximation we arrive at the CAPM discount factor:

Since the time-varying coefficients in (8) ultimately capture time-varying marginal utility growth, I pursue an alternative expression for marginal utility growth used to drive the time variability of a0t and a1t using a discrete time general equilibrium macroeconomic growth model. The model; The model developed here follows the closed economy exogenous growth model without government of King and Rebelo (1999) and is illustrated in Figure 1. To ensure the existence of a steady-state, exogenous growth is introduced via labor augmentation consistent with Sala-i

The Cobb-Douglas production function was chosen for several reasons. First, by construction this production function exhibits constant returns to scale, consistent with the empirical findings of Jorgenson (1972). Second, the empirical evidence noted by Jorgenson (1972) suggests the estimated elasticity of substitution for the constant elasticity of substitution (CES) production function is not significantly different from unity and therefore the CES reduces to CobbDouglas form. Third, Arroyo (1996) suggests the Cobb-Douglas form is probably more descriptive of aggregate technological conditions. Maximization problem Following the model of King and Rebelo (1999), all trending variables are scaled by Xt and transformed variables are denoted by lower case letters (e.g., yt = Yt/Xt). The infinitely-lived central planner maximizes discounted expected utility:

This expression provides a readily observable proxy for unobservable marginal utility growth that avoids the theoretically troublesome utility function specification and empirically troublesome consumption data. Now I proceed towards connecting with the time-varying discount factor coefficients a0t and a1t.
4 Empirical construction and characteristics of

with Ky;q equal to capital for quarter q in year y, Iy;i equal to investment for quarter i in year y, and Ky equal to capital at the end of year y. Following the parametrization of King and Rebelo (1999), the value for the rate of time preference b is set to 0.984, the labor share of output _ is set to 2=3, the quarterly growth rate is set to 1.004, and the quarterly depreciation rate _ is set to 0.025. Given this parametrization and current dataset, E [] = 1:0140, [] = 0:0030, and [] =E [] = 0:0029.
5 [Un]conditional estimation using

under the assumption that the relevant information is captured by . Cochrane (2005) noted that an unconditional Euler condition is implied when using a discount factor with constant coefficients, as in equation (7). Thus, if there were some way to express the discount factor with time-varying coefficients, equation (8), in a form with constant coefficients, an unconditional Euler condition could also be used. To do so, begin by defining a0t and a1t as linear functions of t:

Given the relatively high volatility of RW, the stochastic discount factor (7) has an inherent advantage over the low volatility consumption growth derived discount factor (4). However, the inability of CAPM to explain anomalies such as the size premium, the value premium, and momentum profits is indicative of model mis-specification. Cochrane (2005) noted that if the discount factor (7) is to

where bt < 0 is the rate of time preference, subject to several constraints. First, all output is either consumed or invested:

The first order conditions (17) and (18) can be combined to reveal the key contribution of this study: an alternative proxy for marginal utility growth
price multiple assets simultaneously, the coefficients can not be constant and the correct discount factor representation is Martin (1990). As such, the production function can be specified in CobbDouglas form as

In addition, capital stock evolves according to the perpetual inventory method:

Table 1 summarizes the sources, frequency, and availability of the data used to construct . Aggregate macroeconomic data are obtained from the Bureau of Economic Analysis National Income and Product Accounts tables (BEA NIPA tables). Using the BEA NIPA aggregate macroeconomic data, data, output (Y) is defined as gross domestic product (GDP), capital (K) is defined as private non-residential fixed assets in place, investment (I) is defined as private non-residential fixed investment, and consumption (C) is defined as non-durable consumption. All nominal data are converted to real using the CPI deflator from the U.S. Bureau of Labor Statistics (BLS). Unfortunately quarterly capital data are unavailable from the Bureau of Economic Analysis therefore the series must be estimated. The quarterly capital series is constructed from annual capital data and quarterly investment data following the procedure of Balvers and Huang (2007). Capital in quarter q is computed as:

Where is the rate of depreciation. Noting yt = Atkt 1-a and combining constraints (13) and (14):

The low volatility and coefficient of variation of render it unsuitable for use as a stand-alone discount factor since it will not satisfy the Hansen- Jagannathan bounds. However, the low volatility may be beneficial in a conditional estimation from two perspectives. First, from a scaled payoff perspective, a low volatility instrument is representative of a feasible trading strategy due to low transactional activity and costs. Second, from a scaled discount factor perspective, when the coefficients of discount factor (8) are allowed to vary with , they capture changes in risk aversion around unity. If we allow a0t and a1t to be functions of t, the conditioning information in the original Euler condition can be stated explicitly:

Inserting these coefficients into (8) yields:

In other words, a conditional estimation (conditioned on t) with a single factor (Rtw+1) is equivalent to an unconditional estimation with three factors (t, Rtw+1 and tRtw+1): E[mt+1 Rt+1 |t] = 1 E [m*t+1 Rt+1] = 1 where mt+1 is represented by equation (7) and m*t+1 is represented by equation (24).
6 Theoretical evidence of refined discount factor efficacy

the Lagrangian is:

Where a0t and a1t are time-varying

where Yt is GDP, Kt is capital input, Nt is labor input, At is the random produc-

Of course, whenever one discusses instruments they are exposed to the criticism of omitting instruments relevant to the estimation. However, given the equality between the instrument and marginal utility growth, I proceed

Here I show that the coefficient of variation (CV) of the refined discount factor with time-varying coefficients (m*t+1 ) is likely to be greater than that of the original discount factor with fixed coefficients (mt+1). The larger CV suggests a greater likelihood of satisfying the Hansen-Jagannathan bounds. For notational simplicity, let X = t and >>>

FSR Forum | september 2009 | 10

FSR Forum | september 2009 | 11

Y = Rtw+1 . Dropping subscriptsand rewriting the two discount factors: Computing the expected values of m and m*: It is quite common in financial literature to examine excess returns: Rte = Rt Rft where Rf represents the risk free rate. The Euler condition to excess returns is Et [mt+1 Ret+1] = 0 Note that in this form E [(2m)Re] = 0 is also true. As such, the coefficients of m are not identified. Therefore some form of normalization must be chosen. Let a0 = 1 and a00 = 1 in equations (25) and (26), respectively:

larger numerator of CV [m*].

7 Conclusion

Recall that X is the marginal utility growth proxy and Y is the equity market return. Since marginal utility growth is positively correlated with equity returns we know cov [X; Y ] > 0 and therefore: E [m*] < E [m] Half of the battle is finished. We now know that, all else constant, the smaller denominator of

The macroeconomic growth model yields a productivity-based proxy for marginal utility growth that bypasses two key drawbacks of traditional marginal utility growth proxies. First, the difficulty in obtaining explicit utility functions, which are unobservable, is bypassed by using an observable productivitybased proxy. Second, the more reliably measured production data avoids the problem measurement error associated with consumption data. A refined discount factor was constructed using the necessarily timevarying coefficients

Although the choice of normalization seems arbitrary, Cochrane (2005) noted that the choice is purely one of convenience and the pricing errors are independent of the choice of normalization. Another simplification is to constrain the expected value of the coefficients of (28) to the values in (27). Beginning with the coefficient on Y :

The macroeconomic growth model yields a productivity-based proxy for marginal utility growth that bypasses two key drawbacks of traditional marginal utility growth proxies
CV [m*] will translate into a larger CV than m. Now lets proceed to the numerator. The variance of m and m* can be computed as: expressed as linear functions of the new marginal utility growth proxy. The new discount factor was shown to map directly into an unconditional estimation framework. In particular, the time-varying coefficients of the refined discount factor account for time-varying expected returns, variances, covariances, risk free rates, and risk aversion. Theoretical evidence was provided to demonstrate the greater likelihood of the refined discount factor to satisfy the Hansen-Jagannathan bounds. As such, the refined discount factor with time-varying coefficients could price those assets that the standard CAPM discount factor with fixed coefficients can not. |||

Proceeding to the intercept term:

The last equation reveals a01 = 0 since E [X] > 0. The static CAPM and dynamic CAPM discount factors with the unity intercept normalization and constrained coefficients are summarized below:

Now the goal is to demonstrate the coefficient of variation for m* is greater than that of m:

Given the complexity of expanding var [XY ] and cov [Y,XY ] it is difficult to rigorously prove inequality (33). However, we do know that var [XY ] > 0 and it can be shown that cov [Y,XY ] > 0 (Moore, 2009). Thus it is the empirical researchers task to arrive at coefficients a10 and a11 such that inequality (33) holds. The task does not appear to be monumental given the unambiguously smaller denominator and quite likely

Nu lees je wel eens over staatsbedrijven. Binnenkort mag jij ze beheren.

Het Rijk is mede-eigenaar van 35 grote ondernemingen. Het ministerie van Financin is namens de Staat aandeelhouder en verantwoordelijk voor bijvoorbeeld de verkoop van het vervoersbedrijf Connexxion. Maar ook de oprichting van de onderneming die de Zuidas moet gaan ontwikkelen of de aankoop van een aandelenbelang in het Rotterdamse Havenbedrijf. Op dit ministerie werk

Financin zoekt startende bedrijfseconomen

Wij bieden je van meet af aan veel ruimte voor eigen verantwoordelijkheid. Het kan dan ook zomaar gebeuren dat je direct na je studie staatsbedrijven mag beheren. Dat moet je willen, dat moet je kunnen. Iets voor jou? Toptalent is van harte welkom. Zeker als je binnenkort als bedrijfseconoom afstudeert. Kijk voor meer informatie op Je sollicitatie mail je naar of je belt 070-3428532.


Arroyo, Cristino R., 1996, Testing a production-based assetpricing model, Empirical Inquiry pp. 357377. Balvers, Ronald J., and Dayong Huang, 2007, Productivity- based asset pricing: Theory and evidence, Journal of Financial Economics 86, 405445. Cochrane, John H., 2005, Asset Pricing (Princeton University Press). Hansen, Lars P., and Ravi Jagannathan, 1991, Implications of security market data for models of dynamic economies, Journal of Political Economy 99, 225262. Jorgenson, Dale W., 1972, Investment behavior and the production function, Bell Journal of Economics 3, 220251. King, Robert G., and Sergio T. Rebelo, 1999, Resuscitating real

business cycles, in John B. Taylor, and Michael Woodford, ed.: Handbook of Macroeconomics (Amsterdam: North Holland). Lucas, Jr., Robert E., 1978, Asset prices in an exchange economy, Econometrica 46, 142614. Mehra, Rajnish, 2003, The equity premium: Why is it a puzzle?, Financial Analysts Journal pp. 5469., and Edward Prescott, 1985, The equity premium: A puzzle, Journal of Monetary Economics 15, 145161. Moore, David J., 2009, Useful statistics properties, http://djmphd. Sala-i Martin, Xavier, 1990, Lecture notes on economic growth, Working Paper No. 3563.

je altijd aan uitdagende projecten met grote maatschappelijke gevolgen. Ook als starter, want je doet direct mee als volwaardig teamlid. Dit betekent wel dat wij veel van jou verwachten. Bij Financin tel je meteen mee.